Drought as one of the natural phenomena and one of the greatest climate problems has been of central importance and investigated as a matter of necessity. In recent decades, Iran and especially Ilam province faced drought problem and this research was conducted with the aim of predicting drought in Ilam (semi-arid region) and Dehloran (arid area) stations, as well as evaluating the efficiency of the neural network. In this study, the data of temperature, precipitation, relative humidity, wind speed, and monthly sunny hours in the period of 1983–2013 were used to calculate the standardized precipitation index (SPI) and moving average of 3, 5, and 7 years. Also, artificial neural network was employed for monitoring the drought in Ilam (semi-arid region) and Dehloran (arid region) stations. Accordingly, the artificial neural network (ANN) was modeled after determining the best input composition and the best number of data using the backpropagation algorithm and C-Sharp programming language. Then, the possible future drought was predicted in both stations. The architecture of network was optimized in form of 5-30-1, 5 neurons in input layer (precipitation, temperature, relative humidity, wind speed, and sunny hours); 30 neurons in the hidden layer and a neuron in the output layer (SPI) after training and repeated errors. In order to evaluate ANN performance in climate elements simulation, four designed models were investigated using different learning functions and the number of variously hidden neurons and the error was calculated using MAE, RMSE, and R2 indices. The results showed that in the stations, bipolar sigmoid function pattern and updating weights by the method of the moment were introduced as the optimal model, due to the lowest error rate and highest correlation between input and output data. Then, outputs of the program were compared with observed SPI.
